Neuronpedia logo - a computer chip with a rounded viewfinder border around it

    Neuronpedia

    INDEX
    Explanations

    programming safety principles

    np_acts-logits-general · gemini-2.5-flash-lite

    assistant safety-refusal boilerplate: declarations that the AI cannot comply and references to its safety guidelines, ethical principles, and programming by its creators.

    oai_token-act-pair · gpt-5Triggered by @yooniel31

    statements where the assistant refuses a request by citing safety rules, limits, or that it is "programmed" to be safe (i.e., refusal/safety-policy language).

    oai_token-act-pair · gpt-5-miniTriggered by @yooniel31
    New Auto-Interp
    Negative Logits
    ⌛
    0.39
    を使った
    0.38
     huj
    0.38
    igkeits
    0.37
    mouseX
    0.37
     suasana
    0.37
    成了
    0.37
     combos
    0.37
     eyeing
    0.36
     potions
    0.36
    POSITIVE LOGITS
     programmed
    1.59
     programming
    1.48
    programmed
    1.38
     Programming
    1.31
    Programming
    1.30
    programming
    1.26
     programmers
    1.23
     programmer
    1.20
     программи
    1.20
     programación
    1.14
    Act Density 0.170%

    No Known Activations